[nodestory] add missing Types file

parent 2dc161cf
Pipeline #5615 passed with stages
in 126 minutes and 24 seconds
{-|
Module : Gargantext.Core.NodeStory.Types
Description : Node API generation
Copyright : (c) CNRS, 2017-Present
License : AGPL + CECILL v3
Maintainer : team@gargantext.org
Stability : experimental
Portability : POSIX
-}
{-# LANGUAGE Arrows #-}
{-# LANGUAGE ConstraintKinds #-}
{-# LANGUAGE QuasiQuotes #-}
{-# LANGUAGE TemplateHaskell #-}
module Gargantext.Core.NodeStory.Types
( HasNodeStory
, HasNodeStoryEnv
, hasNodeStory
, HasNodeStoryImmediateSaver
, hasNodeStoryImmediateSaver
, HasNodeArchiveStoryImmediateSaver
, hasNodeArchiveStoryImmediateSaver
, NodeStory(..)
, NgramsState'
, NgramsStatePatch'
, NodeListStory
, ArchiveList
, initNodeListStoryMock
, NodeStoryEnv(..)
, initNodeStory
, nse_getter
, nse_getter_multi
, nse_saver_immediate
, nse_archive_saver_immediate
-- , nse_var
, unNodeStory
, Archive(..)
, initArchive
, archiveAdvance
, unionArchives
, a_history
, a_state
, a_version
, combineState
, ArchiveStateSet
, ArchiveStateList )
where
import Codec.Serialise.Class
import Control.Lens (makeLenses, Getter, (^.))
import Control.Monad.Except
import Control.Monad.Reader
import Data.Aeson hiding ((.=), decode)
import Data.Map.Strict qualified as Map
import Data.Monoid
import Data.Profunctor.Product.TH (makeAdaptorAndInstance)
import Data.Semigroup
import Data.Set qualified as Set
import Database.PostgreSQL.Simple.FromField (FromField(fromField), fromJSONField)
import Gargantext.API.Ngrams.Types
import Gargantext.Core.Types (NodeId(..))
import Gargantext.Core.Utils.Prefix (unPrefix)
import Gargantext.Database.Admin.Config ()
import Gargantext.Database.Prelude (DbCmd')
import Gargantext.Database.Query.Table.Ngrams qualified as TableNgrams
import Gargantext.Database.Query.Table.Node.Error (HasNodeError())
import Gargantext.Prelude hiding (to)
import Opaleye (DefaultFromField(..), SqlJsonb, fromPGSFromField)
------------------------------------------------------------------------
{- | Node Story for each NodeType where the Key of the Map is NodeId
TODO : generalize for any NodeType, let's start with NodeList which
is implemented already
-}
newtype NodeStory s p = NodeStory { _unNodeStory :: Map NodeId (Archive s p) }
deriving (Generic, Show, Eq)
instance (FromJSON s, FromJSON p) => FromJSON (NodeStory s p)
instance (ToJSON s, ToJSON p) => ToJSON (NodeStory s p)
instance (Serialise s, Serialise p) => Serialise (NodeStory s p)
data Archive s p = Archive
{ _a_version :: !Version
, _a_state :: !s
, _a_history :: ![p]
-- first patch in the list is the most recent
-- We use `take` in `commitStatePatch`, that's why.
-- History is immutable, we just insert things on top of existing
-- list.
-- We don't need to store the whole history in memory, this
-- structure holds only recent history, the one that will be
-- inserted to the DB.
}
deriving (Generic, Show, Eq)
instance (Serialise s, Serialise p) => Serialise (Archive s p)
type NodeListStory = NodeStory NgramsState' NgramsStatePatch'
-- NOTE: 'type NgramsTableMap = Map NgramsTerm NgramsRepoElement'
type NgramsState' = Map TableNgrams.NgramsType NgramsTableMap
type NgramsStatePatch' = PatchMap TableNgrams.NgramsType NgramsTablePatch
-- instance Serialise NgramsStatePatch'
instance FromField (Archive NgramsState' NgramsStatePatch')
where
fromField = fromJSONField
instance DefaultFromField SqlJsonb (Archive NgramsState' NgramsStatePatch')
where
defaultFromField = fromPGSFromField
-- | Combine `NgramsState'`. This is because the structure is (Map
-- NgramsType (Map ...)) and the default `(<>)` operator is
-- left-biased
-- (https://hackage.haskell.org/package/containers-0.6.6/docs/Data-Map-Internal.html#v:union)
combineState :: NgramsState' -> NgramsState' -> NgramsState'
combineState = Map.unionWith (<>)
-- This is not a typical Semigroup instance. The state is not
-- appended, instead it is replaced with the second entry. This is
-- because state changes with each version. We have to take into
-- account the removal of terms as well.
-- instance (Semigroup s, Semigroup p) => Semigroup (Archive s p) where
-- (<>) (Archive { _a_history = p }) (Archive { _a_version = v'
-- , _a_state = s'
-- , _a_history = p' }) =
-- Archive { _a_version = v'
-- , _a_state = s'
-- , _a_history = p' <> p }
-- instance (Monoid s, Semigroup p) => Monoid (Archive s p) where
-- mempty = Archive { _a_version = 0
-- , _a_state = mempty
-- , _a_history = [] }
instance (FromJSON s, FromJSON p) => FromJSON (Archive s p) where
parseJSON = genericParseJSON $ unPrefix "_a_"
instance (ToJSON s, ToJSON p) => ToJSON (Archive s p) where
toJSON = genericToJSON $ unPrefix "_a_"
toEncoding = genericToEncoding $ unPrefix "_a_"
-- | This is the normal way to update archive state, bumping the
-- version and history. Resulting state is taken directly from new
-- archive, omitting old archive completely.
archiveAdvance :: (Semigroup s, Semigroup p) => Archive s p -> Archive s p -> Archive s p
archiveAdvance aOld aNew = aNew { _a_history = _a_history aNew <> _a_history aOld }
-- | This is to merge archive states.
unionArchives :: (Semigroup s, Semigroup p) => Archive s p -> Archive s p -> Archive s p
unionArchives aOld aNew = aNew { _a_state = _a_state aOld <> _a_state aNew
, _a_history = _a_history aNew <> _a_history aOld }
------------------------------------------------------------------------
initNodeStory :: (Monoid s, Semigroup p) => NodeId -> NodeStory s p
initNodeStory ni = NodeStory $ Map.singleton ni initArchive
initArchive :: (Monoid s, Semigroup p) => Archive s p
initArchive = Archive { _a_version = 0
, _a_state = mempty
, _a_history = [] }
initNodeListStoryMock :: NodeListStory
initNodeListStoryMock = NodeStory $ Map.singleton nodeListId archive
where
nodeListId = 0
archive = Archive { _a_version = 0
, _a_state = ngramsTableMap
, _a_history = [] }
ngramsTableMap = Map.singleton TableNgrams.NgramsTerms
$ Map.fromList
[ (n ^. ne_ngrams, ngramsElementToRepo n)
| n <- mockTable ^. _NgramsTable
]
----------------------------------------------------------------------
data NodeStoryPoly nid v ngtid ngid nre =
NodeStoryDB { node_id :: !nid
, version :: !v
, ngrams_type_id :: !ngtid
, ngrams_id :: !ngid
, ngrams_repo_element :: !nre }
deriving (Eq)
data NodeStoryArchivePoly nid a =
NodeStoryArchiveDB { a_node_id :: !nid
, archive :: !a }
deriving (Eq)
$(makeAdaptorAndInstance "pNodeStory" ''NodeStoryPoly)
$(makeAdaptorAndInstance "pNodeArchiveStory" ''NodeStoryArchivePoly)
-- type NodeStoryWrite = NodeStoryPoly (Column SqlInt4) (Column SqlInt4) (Column SqlInt4) (Column SqlInt4) (Column SqlJsonb)
-- type NodeStoryRead = NodeStoryPoly (Column SqlInt4) (Column SqlInt4) (Column SqlInt4) (Column SqlInt4) (Column SqlJsonb)
-- type NodeStoryArchiveWrite = NodeStoryArchivePoly (Column SqlInt4) (Column SqlJsonb)
-- type NodeStoryArchiveRead = NodeStoryArchivePoly (Column SqlInt4) (Column SqlJsonb)
type ArchiveList = Archive NgramsState' NgramsStatePatch'
------------------------------------------------------------------------
data NodeStoryEnv = NodeStoryEnv
{ _nse_saver_immediate :: !(NodeId -> ArchiveList -> IO ())
, _nse_archive_saver_immediate :: !(NodeId -> ArchiveList -> IO ArchiveList)
, _nse_getter :: !(NodeId -> IO ArchiveList)
, _nse_getter_multi :: !([NodeId] -> IO NodeListStory)
--, _nse_cleaner :: !(IO ()) -- every 12 hours: cleans the repos of unused NodeStories
-- , _nse_lock :: !FileLock -- TODO (it depends on the option: if with database or file only)
}
deriving (Generic)
type HasNodeStory env err m = ( DbCmd' env err m
, MonadReader env m
, MonadError err m
, HasNodeStoryEnv env
, HasNodeError err
)
class (HasNodeStoryImmediateSaver env)
=> HasNodeStoryEnv env where
hasNodeStory :: Getter env NodeStoryEnv
class HasNodeStoryImmediateSaver env where
hasNodeStoryImmediateSaver :: Getter env (NodeId -> ArchiveList -> IO ())
class HasNodeArchiveStoryImmediateSaver env where
hasNodeArchiveStoryImmediateSaver :: Getter env (NodeId -> ArchiveList -> IO ArchiveList)
type ArchiveStateList = [(TableNgrams.NgramsType, NgramsTerm, NgramsRepoElement)]
type ArchiveStateSet = Set.Set (TableNgrams.NgramsType, NgramsTerm)
------------------------------------------------------------------------
------------------------------------------------------------------------
-- | Lenses at the bottom of the file because Template Haskell would reorder order of execution in others cases
makeLenses ''NodeStoryEnv
makeLenses ''NodeStory
makeLenses ''Archive
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment